Could this be done for the ternary 2B version
Appreciate your contribution. Could this be done for the 2B version, it does not seem compatible on first try.
2026-07-15 15:06:41,927 - omlx.engine_pool - INFO - [-] - Loading model: Ternary-Bonsai-27B-mlx-2bit
2026-07-15 15:06:41,927 - omlx.engine_pool - INFO - [-] - DFlash enabled for Ternary-Bonsai-27B-mlx-2bit, draft=/Users/jpargac/.omlx/models/peppi314/Bonsai-27B-dSpark-MLX-4bit
2026-07-15 15:06:42,862 - omlx.engine_pool - WARNING - [-] - DFlash start failed for Ternary-Bonsai-27B-mlx-2bit: DFlashDraftModelArgs.init() missing 3 required positional arguments: 'max_position_embeddings', 'tie_word_embeddings', and 'num_target_layers'. Falling back to vlm engine.
2026-07-15 15:06:42,862 - omlx.patches.dflash_lifecycle - INFO - [-] - dflash class patches restored on 2 class(es)
2026-07-15 15:06:42,862 - omlx.engine.dflash - INFO - [-] - DFlashEngine stopped
Or if you could give some guidance on how to do this, I would be interested to learn more.
It should be possible. Still I think your problem is caused by how omlx deals with the drafter. See the errors. You might try to get it work on the 1bit just to make sure. I am still testing the 1bit with this drafter and still trying to figure out if spec-decode on my mac is worth the pain but it does not work out of the box. I also might try on 2bit but later. no guarantee when.
True. Could test that first of course. Good call. Thank you